Dissonance and Meaning Maintenance running head: Dissonance causes compensatory affirmation

Study 2

Participants and procedure

Students completed the same dissonance paradigm online as in Study 1, in exchange for partial credit in a psychology class at the University of Toronto Scarborough. Instead of a scenario about norm or law violations, participants were asked about their belief in God and the role of God in creating and maintaining the world. Specifically, they were asked how strongly they would endorse the following items: “I believe in God”, “I think that God is in control of the events in our universe,” “I think that the actions of God explain what happens in our world,” and “I think that God created all life on the planet.”

Two hundred and twenty-four students participated (72% women, mean age = 19.1, SD = 2.03). The ethnicity of the sample was primarily of South Asian (34.6%), East Asian (29.5%), Western European (12.4%) and other ethnicities (23.7%).

Results

Twenty participants were removed for completing the study too quickly (less than 7 minutes) or taking three standard deviations longer than the average to complete. This study did not include an Oppenheimer-type manipulation check. In the dissonance condition, 42 participants refused to write that the paragraph was interesting (61% compliance), with 12 participants in the control condition refusing (87.5% compliance). As with the previous studies, we include non-compliers and moderate the experiment by willingness to comply, including age and sex as covariates. Additionally, despite random assignment there were more self-described atheists in the dissonance condition (N_Cont = 9; N_Diss = 17); a group that strongly predicts low endorsement on the belief in God scale (d = 1.09, p < .0001) and is unlikely to affirm belief. To address this imbalance, we additionally control for self-reported atheism for our key analysis on belief in God.

The manipulation check indicated that the dissonance paradigm had been successful. Participants in the dissonant condition claimed they had more choice to write that the article was interesting p < .0001, d = .62 CI_.975[.34, .90]. In this study we did not include a measure of direct dissonance reduction. Looking at our key dependent variable (belief in God), dissonance does not cause a significant change in belief as a main effect, M_Diss = 3.29, SD = 1.20, M_Cont = 3.30, SD = 1.21, B = .11, p = .48, d = .1 CI_.975[-.18, .39] (see Table 1), but there was again a significant interaction when willingness to comply was entered as a moderating term, B = .75, p = .054. When participants complied with instructions, those in the dissonance condition reported marginally higher belief in God relative to the control, M_Diss = 3.41, SD = 1.20, M_Cont = 3.27, SD = 1.21; B = .29, p = .098, d = .24 CI_.975[-.05, .52]. For those who didn't comply, the control group again had a stronger indication of affirmation than the dissonant group, although not significantly so, M_Diss = 3.10, SD = 1.20, M_Cont = 3.5, SD = 1.25; B = -.46, p = .18, d = .19 CI_.975[-.09, .47]. Despite this effect not being significant, it is worth noting it is in the same direction and of a similar magnitude to the affirmation effect for non-compliers in Study 1.

In Study 3, we attempted to replicate the cultural affirmation effect using a different dissonance paradigm: arguing in favor of increased tuition. We also attempted to extend the fluid compensation findings by showing that dissonance leads to increased abstraction, as well as affirmation. Recent work in areas of uncertainty, meaninglessness and lack of control show that violations increase a person’s motivation for identifying patterns in the environment. This includes being more likely to perceive illusory visual patterns amongst noise, but can also increase the motivation to see behaviors amongst people as connected, such as with conspiracy theories (Whitson & Galinsky, 2008). Some studies have found that when actual patterns are present, participants show not only increased motivation to detect patterns, but increased ability as well (Proulx & Heine, 2009; Randles et al., 2011).

Study 3

Participants and Procedure

A total of 212 students (73% women) participated in exchange for partial credit in their psychology classes at The University of British Columbia. We used the classic dissonance paradigm (Pittman, 1975; Steele et al., 1981), where students were either instructed, or politely asked, to write in favor of a tuition increase at their university. Participants were 40% East Asian, 34% Western European, 8% South Asian and 18% other or mixed ethnicities.

The study was advertised as soliciting student opinions on university policy. After completing basic demographic materials, participants were told that the university Board of Governors was soliciting student opinions on whether tuition should be raised by 20% for the following academic year. Students in the control condition were told that they had been selected to write in favor of the increase, and to offer valid reasons for it. In the dissonance condition, the researcher explained to the participant that many people had written against the idea, and the researcher would appreciate if they could write a paragraph in favor of an increase. After participants submitted their paragraph, they completed the Positive and Negative Affect Schedule (Watson, Clark, & Tellegen, 1988) as a delay and distractor between the manipulation and dependent variable. It has become common practice to use the PANAS as a filler task for manipulations of uncertainty or meaning, and doing so appears to increase the effect size of the manipulation (Burke et al., 2010). As this measure almost never shows a change following manipulations (e.g. Landau et al., 2006; Proulx & Heine, 2008; Randles et al., 2011), we did not expect to see significant differences between conditions in either self-reported positive or negative affect. This is not to suggest that there is no arousal component to manipulations of uncertainty or dissonance, but that the PANAS is typically insensitive to it. This study was run prior to recent work pointing towards new approaches to measuring affect that may be more sensitive to this type of experience (Lambert et al., 2014; Spunt, Lieberman, Cohen, & Eisenberger, 2012).

Participants were then given an implicit grammar task previously used in other meaning violation studies (Proulx & Heine, 2009; Randles et al., 2011). In this measure, participants are asked to copy a series of letter strings (e.g. XXRVTM), with no additional instructions. Every string adheres to the same strict syntactical grammar (e.g. M can follow X, M, or V, but not T, unless T is preceded by V.; see Dienes & Scott, 2005 for full description). Once completed, they were presented with new strings one at a time on a computer, half of which were based on the same grammar, and half were not. Participants were told to identify the strings that adhered to the grammar of the previously-copied strings, and were not told how many of the new strings were correct. Participants were scored for how many strings they selected (a measure of their motivation to detect patterns) and their actual success on the task. Given past work, we anticipated that those in the dissonant group would both select a greater number of strings, but also be more accurate in selecting strings that correctly matched the previously viewed pattern.

Following the grammar task, participants then completed the social judgment survey dependent variable, as in Study 1, and a measure of dissonance reduction where we asked participants to tell us how they really felt about tuition increases, regardless of what they had written previously. The mean of the items “I support an increase in tuition,” “I think there are many valid reasons why tuition should be increased,” and “students can handle an increase in tuition” were used as a score to measure dissonance reduction. Afterwards, participants were fully debriefed.

Results

As this was an in-lab study, we did not monitor participants for taking an unusually short or long time to complete the study. Seven participants were removed because of procedural error in the study, or because the participant had been in a similar study, leaving 205 participants. Nineteen participants in the forced choice condition (81% compliance) and 46 participants in the dissonance condition (55% compliance) refused to write in favor of a tuition increase. The manipulation check indicated that the dissonance paradigm had been successful. Participants in the dissonant condition claimed they had more choice to write their essay, M = 6.18, SD = 2.63, compared to the control, M = 2.96, SD = 2.62; B = 3.20 p < .0001, d = 1.23 CI_.975[.95, 1.51].

As with Study 1, the main effect for dissonance is significant, M_Diss = 2.86, SD = 1.33, M_Cont = 2.54, SD = 1.23; B = .64, p = .0002, d = .54 CI_.975[.26, .82]. Also as with the previous studies, a significant interaction emerged when compliance was included as a moderator of condition, B = 1.19, p = .0008. Amongst those who complied with instructions, we observed a classic dissonance effect, such that participants in the dissonance condition more strongly supported an increase in tuition, M = 3.51, SD = 1.17, compared to those in the forced choice condition, M = 2.30, SD = 1.07; B = 1.24, p < .0001, d = .96 CI_.975[.67, 1.24]; see Table 1. Those who did not comply showed no difference between conditions, M_Diss = 2.05, SD = 1.04; M_Cont = 1.90, SD = .91; B = .05 p = .87, d = .02 CI_.975[-.30, .26].

We used the same analyses to assess whether participants showed increased motivation and success at detecting implicitly learned patterns; again the main effect of condition was significant, M_Diss = 23.333, SD = 10.27; M_Cont = 20.41, SD = 10.48; B = 2.86 p = .052, d = .28 CI_.975[.0, .56] (see Table 1), with a significant interaction between compliance and condition, B = 8.32, p = .014. Consistent with our hypothesis, the dissonant group who agreed to write the essay showed increased motivation to identify patterns amongst noise B = 5.93, p = .001, d = .48 CI_.975[.20, .77], with the dissonant group selecting a mean of 25.98 strings (SD = 9.93) compared to the control, M = 19.99, SD = 10.11. For those who refused to comply, the control group showed increased motivation, though this results was not significant, M_Diss = 20.11, SD = 9.84; M_Cont = 22.21, SD = 12.011; B = 2.38, p = .40, d = .12 CI_.975[-.16, .40].

Beyond increasing motivation, we anticipated that the dissonant group would actually perform better on the task. This was assessed using two approaches. The first is a measure of overall success, assigning a score based on correct hits minus false alarms. This approach however, confounds response bias with actual implicit learning, because randomly selecting is progressively less likely to lead to a hit with each successful selection. To address this, we also established participant sensitivity using d' (not to be confused with Cohen's d; Brophy, 1986). This index is based on signal-detection theory, and attenuates the response-bias confound by giving relatively stronger weight to each successive hit and false alarm. Using either approach, we did not find a main effect,, though both overall success M_Diss = 7.12, SD = 6.12; M_Cont = 6.21, SD = 5.63; B = .92 p = .27, d = .16 CI_.975[-.44, .12], and sensitivity M_Diss = .80, SD = .69; M_Cont = .76, SD = .71; B = .04 p = .685, d = .06 CI_.975[-.34, .22], were in the predicted direction of dissonance leading to improved performance. We also did not find an interaction between condition and willingness to comply (p = .76 for success and p = .80 for sensitivity). This suggests that the participants were more motivated to identify patterns regardless of whether a signal was present or absent (Whitson and Galinsky, 2008), but were not actually more effective at identifying patterns, as was seen with past meaning violation studies (Proulx & Heine, 2009; Randles et al., 2011).

Fifteen participants were familiar with the social judgment survey from having participated in previous studies, and so were removed only from this analysis. The main effect for condition was significant, such that the dissonant group showed greater affirmation, M_Diss = 503.58, SD = 276.49, M_Cont = 411.31, SD = 265.91; B = 93.87, p = .02, d = .35 CI_.975[.06, .65]. As in Study 1 and 2, there was a moderated effect on affirmation between condition and compliance, B = 216.19, p = .018. Amongst those who complied, higher fines were assigned for the prostitution scenario in the dissonance condition, M = 539.21, SD = 245.03, compared to the control condition, M = 387.01, SD = 243.36; B = 157.24, p = .002, d = .48 CI_.975[.19, .78] . There was no significant effect for those who did not comply, though as with Studies 1 and 2, the control group numerically showed a greater tendency to affirm, M_Diss = 458.4, SD = 309.09, M_Cont = 509.83, SD = 332.70; B =58.65, p = .44, d = .12 CI_.975[-.18, .41].

Although we anticipated no difference between conditions on the PANAS, the dissonant group reported significantly more positive affect, M_Diss = 2.57, SD_Diss = .57; M_Cont = 2.41, SD_Cont = .613; B = .16, p = .054 , d = .27 CI_.975[0.00, .57]. Negative affect was not significantly different, M_Diss = 1.86, SD_Diss = .494; M_Cont = 1.79, SD_Cont = .45; B = .07, p = .32, d = .01 CI_.975[-.14, .42]) relative to the control group. Controlling for positive and negative affect does not change the effect of dissonance on implicit grammar motivation, or compensatory affirmation.

It is worth noting that, although this study did not explicitly assess order effects, as order was not manipulated, we observed classic dissonance reduction following participants' affirmations. The implication is that the affirmation has not significantly attenuated the motivation to reduce dissonance. This is a replication of what we observed in Study 1 and is considered in the general discussion. We also ran follow-up analyses to test whether those who showed high motivation on the grammar task were also the ones to affirm the bond, but the two effects appear to be independent. The correlation between motivation to identify strings and affirmation is r = .01, p = .9 and controlling for one of the variables does not meaningfully change the effect of dissonance on the other. One implication is that there may be distinct individual differences regarding how strongly someone feels the motivation to see new patterns vs. explicitly affirm an important belief following dissonance.

In Studies 1-3, we showed that classic dissonance manipulations can lead to compensatory affirmation as well as attempts to reduce dissonant cognitions. In Study 4, we attempt to show parallel effects of a dissonance manipulation and a meaning violation.

Study 4

Participants and procedure

Two hundred and forty-two participants were recruited from Tilburg University, (71% women; mean age = 20.8 years, SD = 2.46 years) in exchange for partial credit in a psychology class. All participants were Western European nationals who spoke English and Dutch as a first or second language. All materials were presented in Dutch, except for the boring paragraph and video manipulations, which were presented in English.

Participants entered the lab room and were seated in front of a computer. The experiment began with participants filling out a demographics questionnaire. Participants then had their attitudes towards positive discrimination assessed prior to the experimental manipulation. Positive discrimination, another term for affirmative action, refers to efforts towards elevating the status of minority groups to increase their representation in society. We anticipated that experiencing either dissonance or a meaning violation would motivate participants to affirm their recently assessed positive discrimination attitudes in the direction that they already associate with. Participants should produce a main effect for affirmation when most people in the study generally hold similar opinions (such as we’d expect almost all students to be in favor of punishing someone who has broken a law); we would expect that people are in fact affirming the perspective they already hold, rather than shifting towards the same pole on a topic regardless of prior belief (e.g. Ben-Ari, Florian, & Mikulincer, 1999; Kosloff, Greenberg, Weise, & Solomon, 2010; Proulx & Major, 2013; Vess et al., 2009).

Following this, participants engaged in one of 3 experimental conditions. Participants either completed a neutral (control) task, experienced induced-compliance dissonance using the boring paragraph paradigm from Studies 1 and 2, or were exposed to a surreal video previously used as a meaning violation (Randles et al., 2013). In the control condition, participants were firmly told (without choice) to write that the boring paragraph was interesting, and they then viewed the control version of the video clip (described below). Participants in the dissonance condition completed the induced compliance dissonance manipulation and viewed the control version of the video clip. Participants in the meaning violation condition were also firmly told to write that the article was interesting, identical to the control condition instructions, but they then viewed the surreal version of the video clip. After these videos, participants completed the PANAS and then our dependent variable, an affirmation measure of positive discrimination. Participants were then debriefed and excused from the experiment.

Materials

Pre-manipulation Positive Discrimination Scale. Participants completed a 4-item measure of their relevant attitudes: “I think it is positive that the Dutch government tries to increase the number of women and minority policemen.”, “Women must be given more opportunities, compared to men, to occupy chief executive or general management positions.”, “It is a good thing that the European parliament compensates small countries for their potential lack of influence by giving them more parliament seats per citizen. (E.g., Luxembourg receives 1 seat per 80.000 citizens were Germany receives 1 seat per 800.000 citizens).”, “I think it is a good idea from the Dutch Organization for Scientific Research (NWO) to encourage the promotion of female academics to senior lecturer (or professorial) level.”

Meaning threat and control videos. All participants watched three video clips, under the pretense that they would be asked questions regarding various details later on. This manipulation was previously used in Randles et al. (2013), and has been shown to lead to compensatory affirmation. The first and last clips were the same for both conditions, including a segment from a Disney cartoon starring Donald Duck, and a Peanuts cartoon starring Snoopy. The first clip was intended to help participants get comfortable with the task, while the latter clip added a delay between the manipulation and dependent variable, a practice that has been shown to increase the robustness of meaning violations (e.g., Burke et al., 2010). In the meaning violation condition, participants watched a 4-min clip from the short film, Rabbits, created by David Lynch (2002). The film at first appears to resemble a sitcom, but includes non-sequiturs and a complete lack of narrative, random laugh and applause tracks, and all characters dressed in rabbit costumes with no explanation or reference. The control group watched a clip from The Wizard of Oz. This clip replaced the original control video (Randles et al., 2013) featuring a clip from The Simpsons cartoon show, to reduce potential positive affect as a confounding explanation for the effect.

Positive discrimination affirmation. Our dependent variable was a 1-item measure of support for affirmative action on a 6-point Likert scale: “How do you generally feel about acts, policies, and measures that are driven by the idea of positive discrimination?”

Results

We expected that after experiencing either dissonance or a meaning violation, participants would feel motivated to more strongly affirm elements of their committed social justice worldview relevant to positive discrimination (Proulx & Major, 2013). That is, those who held general attitudes in favor of positive discrimination should make a post-manipulation judgment that endorses positive discrimination more strongly than in the control condition, and those who were relatively opposed to positive discrimination and experienced either manipulation should make a post-manipulation judgment that less strongly endorses positive discrimination, relative to their like-minded participants in the control condition. To test this, initial positive discrimination attitudes scores were mean-centered and entered as a continuous covariate and moderator of condition. Experimental conditions were entered as dummy coded variables referencing the control group. We anticipated that participants would polarize their judgment concerning positive discrimination more strongly following both the meaning violation and dissonance condition, relative to the control. Because almost no participants in the control (N = 1) and meaning violation (N = 4) conditions refused to comply, it is impossible to conduct the analysis with compliance as a moderator. A total of 38 participants in the dissonance condition refused to write that the paragraph was interesting (52% compliance). As with the previous studies, we control for sex and age.

Prior positive discrimination attitudes significantly predicted the post-manipulation positive discrimination judgment in the control condition, B = .31, p = .016. This effect was qualified by a predicted (though marginal) interaction with the dissonance group, B = .32, p = .09, d = .23 CI_.975[-.04, .51] and significant interaction with the meaning violation group, B = .42, p = .022, d = .32 CI_.975[.06, .59], such that both showed a stronger predictive relationship between pre-manipulation positive discrimination attitudes and post-manipulation positive discrimination judgment relative to the control; see Table 1, Figure 1. To observe the relationship between prior attitude/post judgment positive discrimination for each condition, we re-ran the model after centering the condition dummy variables on the meaning violation condition, and then dissonance. This approach allows evaluation of the main effect at different levels of the critical variable (in our case, condition) without inflating Type I error by running separate models (Aiken & West, 1996). For the dissonance group, the slope was B = .63, p < .0001, and for the meaning violation group it was B = .73, p < .0001; see Figure 1. When non-compliers are removed, the interaction between pre/post measure of positive discrimination and control vs. dissonance conditions becomes significant, B = .50, p = .03, d = .38 CI_.975[.04, .73], with the main effect of positive discrimination attitudes increasing to B = .80, p < .0001 for the dissonance condition; The interaction for the meaning violation group reduces slightly and become marginally significant, B = .347 p = .07, d = .32 CI_.975[-.03, .67]

In both cases, whether participants wrote against their beliefs or watched a surreal video, those who already held relatively negative attitudes towards positive discrimination became increasingly disapproving, while those who had relatively positive attitudes towards it further affirmed this belief.

As is typical for meaning violation studies, participants showed no difference in either positive or negative affect, as measured by the PANAS, relative to the control condition, all ps > .17. Given that we did not anticipate significant PANAS results in Study 3 or 4, we are hesitant to interpret the effects in Study 3.